The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests repository source code and uses LLMs to perform automated reviews and generate code fixes, creating an attack surface where malicious instructions embedded in the code could influence agent behavior.
Ingestion points: Processes repository source code identified through feature records and paths changed between the project baseline and the campaign HEAD.
Boundary markers: The instructions do not define explicit delimiters or instructions to the LLM to disregard potentially malicious prompts embedded within the files being reviewed or fixed.
Capability inventory: The skill can execute code via pnpm exec clawpatch fix, run focused project test commands, and perform git commit operations to apply findings.
Sanitization: No sanitization of ingested code content is mentioned before the data is passed to the Codex and Claude model providers.
[COMMAND_EXECUTION]: The skill executes local binaries and Node.js scripts as part of its automated workflow.
Evidence: Invokes pnpm exec clawpatch for status, review, and fix operations. Executes a local Node.js validation script (check-clawpatch-feature-map.mjs) to verify feature map integrity and performs various Git operations.

spine-clawpatch